Understanding the News That Moves Markets

Home , Corn Exchange Bank, Manhattan Company, National Bank of Detroit, New York Trust Company, Texas Commerce Bank

Understanding the News that Moves Markets

EMNLP 2018 November 3, 2018

Gideon Mann Head of Data Science, Office of the CTO My journey: bad hair to no hair What this talk is not about: How to make money on the stock market What this talk is about:

Financial technology The news that moves markets Computer Things BlocksWorld Financial Technology Financial Technology II Financial Technology III

What is the finance ecosystem? 50,000 40,000

S Bonds

U 30,000

Bonds have more market n

i 20,000

l 10,000 value than equities l i Equities B 0 03 04 05 06 07 08 09 10 11 12 13 14

5,000 4,500 4,000

FX has the most daily D

S 3,500

U 3,000

n notational volume i 2,500

n 2,000

l 1,500

l i 1,000

B 500 0 Equities Commodities Bonds FX

Govt + Corp Equities Vastly more structured Municipal Bonds products and derivatives Derivatives than equities Structured Products

Sources: WCAUUS Index, http://www.sifma.org/research/statistics.aspx News moves markets Not all channels are the same…

[Dredze et al. 2016] [Osborne, Dredze 2014] …not all sources are the same

First Bloomberg Headline

New York Times story

SEC Announcement ~12% Broadcomm is said to explore deal to acquire chipmaker Qualcomm Human Language Technology, not Renaissance Technologies Time is not a myth

Days & Hours Fundamental Investor Minutes Institutional Trader Milliseconds Event Driven Algorithmic Trader Microseconds High Frequency Trading Detection Extraction Interpretation Presentation

New text arrives Extract structured Relate new data to Inform the user on data from text existing data what happened Detection Extraction Interpretation Presentation Document ingest and search Twitter: Ranking financial tweets

[Ceccarelli, Nidito, Osborne 2016] A disaster in the White House is huge

Over $136 billion was wiped out in minutes Credibility Twitter: Geolocation credibility

[Preotiuc-Pietro, Guntuku, Ungar 2017] [Preotiuc-Pietro, Liu, Hopkins, Ungar 2017] [Giorgi, Preoţiuc-Pietro, Buffone, Rieman, Ungar Schwartz 2018] [Dredze, Osborne, Kambadur 2016] Detection Extraction Interpretation Presentation Extraction from natural documents Cats all the way down Scatteract

[Cliche, Rosenberg, Madeka, Yee 2017] How Bill O’Reilly leaving Fox fired up O’Reilly Auto Parts stock

[Yang, Irsoy, Rahman 2018] [Tsai, Roth 2018] The Chemical Bank of New York (est. 1823) Citizens National Bank (est. 1851, acq. 1920) Corn Exchange Bank Chemical Bank (est. 1852, acq. 1954) Financial named (reorganized 1988) New York Trust Company (acq. 1959) entity extraction Chemical Bank Texas Commerce Bank (merged 1991) (est. 1866, acq. 1986)

Manufacturers Trust Company Manufacturers Hanover (acq. 1905) Chase Manhattan Bank (merged 1961) (merged 1996) Hanover Bank (est. 1873) Bank of the JPMorgan Chase Manhattan Company (merged 2000) Chase Manhattan Bank (est. 1799) (merged 1955) Chase National Bank of the City of New York (est. 1877) Guaranty Trust Company of New York J.P. Morgan & Co. (est. 1866) (formerly Morgan Guaranty Trust) J.P. Morgan & Co. (merged 1959) (The House of Morgan) JP JPMorgan Chase & Co. (est. 1895) City National Bank Banc One Corp. & Trust Company (merged 1968) Farmers Saving & Trust Company First Chicago Corp. (est. 1863) Bank One First Chicago NBD NBD Bancorp. (acq. 2000) (merged 1995) (formerly National Bank of Detroit) (est. 1933) Louisiana’s First Commerce Corp.

Bear Stearns (est. 1923, acq. 2008) Bear Stearns

Washington Mutual (founded 1889) Great Western Bank (acq. 1997) H.F. Ahmanson & Co. (acq. 1998) Bank of United of Texas (acq. 2001) Dime Bancorp, Inc. (acq. 2002) Providian Financial (acq. 2005) More data, more distant supervision for fact extraction

Starbucks reports 2017 fourth-quarter and three-month trailing results

Three-month trailing results Three-month trailing net revenue of $16,688.53

455 2000 4000 8000 22737 Supervised Distant Supervision

[Meerkamp, Zhou, 2017] Detection Extraction Interpretation Presentation Sanford Bernstein’s Toni Sacconaghi: “And so where specifically will you be in terms of capital requirements?”

Musk “Excuse me. Next. Boring, bonehead questions are not cool. Next?” Constructing portfolio from a statistic: factor loading

1.0

.9 Home Depot 0.9 Long .8 Starbucks 0.87

.1 Yum Brands 0.1 Short McDonalds 0.05 0 Sentiment Alpha: Russell 2000 March 11, 2011: Earthquake in Japan

Toyota Motor Hokkaido Plant

Toyota Motor Tohoku Plant

Kanto Auto Workers Iwate Plant

Cental Motor Miyagi Plant Explaining Knowledge Graph relationships

Starbucks founder of Zev Siegl d a te

fo u n “In 1971, Siegl, Bowker and Baldwin f d o e r d e d f established the Starbucks Coffee

n o

u o r

f e

d Company.”

1971

Gordon Bowker [Stefanoni, Motik, Kostylev 2018] Jerry Baldwin [Abujabal, Roy, Yahya 2018] [Voskarides, Meij, Reinanda, Khaitan, Osborne, Stefanoni, Kambadur, Rijke, 2018] [Voskarides, Meij, Rijke 2017] Detection Extraction Interpretation Presentation Automated news News on demand NL QA / structured extraction from databases

[Arkoudas, Yahya 2018] Detection Extraction Interpretation Presentation

News classification Natural document Event studies Natural language and ranking understanding generation Signal construction Topic identification Chart understanding NL interfaces to Knowledge graph databases Twitter user demographics Information extraction inference

Twitter geolocation Distant supervision Detection Extraction Interpretation Presentation

Twitter geolocation Distant supervision

Speed Precision We’re not in BlocksWorld anymore Lynette Hirschman on Reading comprehension

“A system has learned from a text if, by reading the text, it is able to answer questions it would not have been able to answer before having access to it.” More Info: https://www.TechAtBloomberg.com/NLP

Thank You!